Explainable Cross-Market Causal Discovery via Financial Knowledge Graphs

Abstract

This paper addresses the problem of causal inference in cross-market financial information modeling. It proposes a framework for causal discovery and explainable modeling based on financial knowledge graphs. First, a knowledge graph is constructed, covering various financial entities such as companies, markets, and events, along with their relationships. This enables structured integration and semantic representation of heterogeneous financial data. Then, a graph neural network is applied for embedding learning over the knowledge graph. Combined with structural equation modeling and causal inference techniques, the framework identifies potential causal paths across multiple markets. On this basis, a path attention mechanism and the Shapley value method are further introduced to support causal chain interpretability and the importance assessment of causal factors. Experiments are conducted on publicly available financial datasets. Multiple comparison and ablation experiments are designed to evaluate the model's performance in terms of accuracy, causal effect estimation error, and path coverage. The results demonstrate the effectiveness and superiority of the proposed method in both structural modeling and interpretability.